[Bug c++/102932] New: Wrong implementation of abs with O3

[Bug c++/102932] New: Wrong implementation of abs with O3

[rajpal.gusain at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102932

            Bug ID: 102932
           Summary: Wrong implementation of abs with O3
           Product: gcc
           Version: 8.3.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: rajpal.gusain at gmail dot com
  Target Milestone: ---

Looks like while optimizing abs, optimizer trips and produces wrong result

Test case:

#include <cstdlib>
#include <iostream>
using namespace std;
int main()
{
  int left , right ;
  cout << "Enter left and right\n";
  cin >> left >> right;
  unsigned int rangeSize = abs(left-right)+1;
  cout << rangeSize << endl;
  return 0;
}

Inputs: 0, 2147483647 (2**31-1)

Compile command:
g++ a.cpp -O3

Compiler version:
g++ --version
g++ (GCC) 8.3.0
Copyright (C) 2018 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.


Output:
18446744071562067968

Workaround:
1. Non-optimized compilation works as expected. 
2. Passing fno-builtin-abs to compiler works.
g++ a.cpp -O3 -fno-builtin-abs

Looking at generated assembly, it looks like implementation of abs is flawed
and needs to be fixed.

[Bug c++/102932] Wrong implementation of abs with O3

[xtkoba at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102932

Tee KOBAYASHI <xtkoba at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |xtkoba at gmail dot com

--- Comment #1 from Tee KOBAYASHI <xtkoba at gmail dot com> ---
Signed integer overflow. You can use -fwrapv.

[Bug c++/102932] Wrong implementation of abs with O3

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102932

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |INVALID
             Status|UNCONFIRMED                 |RESOLVED

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
As mentioned signed integer overflow is undefined. You would have seen an error
at runtime if you used -fsanitize=undefined.

[Bug c++/102932] Wrong implementation of abs with O3

[rajpal.gusain at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102932

--- Comment #3 from Rajpal Singh <rajpal.gusain at gmail dot com> ---
(In reply to Tee KOBAYASHI from comment #1)
> Signed integer overflow. You can use -fwrapv.

Thanks ! Yes, it's signed integer overflow, is there any way to catch it
statically at compile time ?

[Bug c++/102932] Wrong implementation of abs with O3

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102932

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Rajpal Singh from comment #3)
> Thanks ! Yes, it's signed integer overflow, is there any way to catch it
> statically at compile time ?

Not really because there would so many false positives, it is not worth the
effort.